Is Perceptual Learning Constrained to Operate Through Perceptual (Not Sensory) Representations?

Abstract

Many models of perceptual learning (PL) have implied that learning mechanisms have direct access to early sensory analyzers. We explored the hypothesis that PL may operate only through interpreted perceptual representations. In most studies, differences in sensory attributes and perceptual representations are correlated. For example, a certain retinal size is perfectly correlated with perceived size if distance information is constant. Thus size learning could involve retinal size, perceived size, or both. Retinal and perceived size may be decorrelated if distance information varies. We decorrelated sensory attributes and perceptual representations in three experiments. In discriminations involving size, brightness, and velocity, we found that when basic sensory inputs were made the sole basis of categorization (with context varying), PL did not occur (e.g., PL did not occur when classification depended on retinal size, with perceived size variable.) PL did occur reliably for comparable learning conditions in all three domains when sensory attributes and perceptual representations were perfectly correlated. In additional experiments, we verified that PL was also possible for classifications based on perceptual equivalence alone (i.e., classification was not possible based on sensory information). PL may be constrained to operate through perceptual representations rather than directly on early sensory encodings. This limitation of PL mechanisms may be beneficial for reducing the size of the search space in PL without overly restricting what can be learned. Learning may be directed at regularities in the world, not regularities in raw sensory information from which the perceived world is derived. If so, learning may act through perceptual representations of the distal rather than the proximal stimulus. This view does not deny that PL involves changes in early sensory analyzers, only that such changes are likely guided by higher-level, interpreted representations.